net/netfilter/nfnetlink.c

   1 /* Netfilter messages via netlink socket. Allows for user space
   2  * protocol helpers and general trouble making from userspace.
   3  *
   4  * (C) 2001 by Jay Schulist <jschlst@samba.org>,
   5  * (C) 2002-2005 by Harald Welte <laforge@gnumonks.org>
   6  * (C) 2005,2007 by Pablo Neira Ayuso <pablo@netfilter.org>
   7  *
   8  * Initial netfilter messages via netlink development funded and
   9  * generally made possible by Network Robots, Inc. (www.networkrobots.com)
  10  *
  11  * Further development of this code funded by Astaro AG (http://www.astaro.com)
  12  *
  13  * This software may be used and distributed according to the terms
  14  * of the GNU General Public License, incorporated herein by reference.
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/types.h>
  19 #include <linux/socket.h>
  20 #include <linux/kernel.h>
  21 #include <linux/string.h>
  22 #include <linux/sockios.h>
  23 #include <linux/net.h>
  24 #include <linux/skbuff.h>
  25 #include <asm/uaccess.h>
  26 #include <net/sock.h>
  27 #include <linux/init.h>
  28
  29 #include <net/netlink.h>
  30 #include <linux/netfilter/nfnetlink.h>
  31
  32 MODULE_LICENSE("GPL");
  33 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
  34 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
  35
  36 static char __initdata nfversion[] = "0.30";
  37
  38 static struct {
  39         struct mutex                            mutex;
  40         const struct nfnetlink_subsystem __rcu  *subsys;
  41 } table[NFNL_SUBSYS_COUNT];
  42
  43 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
  44         [NFNLGRP_CONNTRACK_NEW]         = NFNL_SUBSYS_CTNETLINK,
  45         [NFNLGRP_CONNTRACK_UPDATE]      = NFNL_SUBSYS_CTNETLINK,
  46         [NFNLGRP_CONNTRACK_DESTROY]     = NFNL_SUBSYS_CTNETLINK,
  47         [NFNLGRP_CONNTRACK_EXP_NEW]     = NFNL_SUBSYS_CTNETLINK_EXP,
  48         [NFNLGRP_CONNTRACK_EXP_UPDATE]  = NFNL_SUBSYS_CTNETLINK_EXP,
  49         [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
  50 };
  51
  52 void nfnl_lock(__u8 subsys_id)
  53 {
  54         mutex_lock(&table[subsys_id].mutex);
  55 }
  56 EXPORT_SYMBOL_GPL(nfnl_lock);
  57
  58 void nfnl_unlock(__u8 subsys_id)
  59 {
  60         mutex_unlock(&table[subsys_id].mutex);
  61 }
  62 EXPORT_SYMBOL_GPL(nfnl_unlock);
  63
  64 #ifdef CONFIG_PROVE_LOCKING
  65 int lockdep_nfnl_is_held(u8 subsys_id)
  66 {
  67         return lockdep_is_held(&table[subsys_id].mutex);
  68 }
  69 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
  70 #endif
  71
  72 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
  73 {
  74         nfnl_lock(n->subsys_id);
  75         if (table[n->subsys_id].subsys) {
  76                 nfnl_unlock(n->subsys_id);
  77                 return -EBUSY;
  78         }
  79         rcu_assign_pointer(table[n->subsys_id].subsys, n);
  80         nfnl_unlock(n->subsys_id);
  81
  82         return 0;
  83 }
  84 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
  85
  86 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
  87 {
  88         nfnl_lock(n->subsys_id);
  89         table[n->subsys_id].subsys = NULL;
  90         nfnl_unlock(n->subsys_id);
  91         synchronize_rcu();
  92         return 0;
  93 }
  94 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
  95
  96 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u_int16_t type)
  97 {
  98         u_int8_t subsys_id = NFNL_SUBSYS_ID(type);
  99
 100         if (subsys_id >= NFNL_SUBSYS_COUNT)
 101                 return NULL;
 102
 103         return rcu_dereference(table[subsys_id].subsys);
 104 }
 105
 106 static inline const struct nfnl_callback *
 107 nfnetlink_find_client(u_int16_t type, const struct nfnetlink_subsystem *ss)
 108 {
 109         u_int8_t cb_id = NFNL_MSG_TYPE(type);
 110
 111         if (cb_id >= ss->cb_count)
 112                 return NULL;
 113
 114         return &ss->cb[cb_id];
 115 }
 116
 117 int nfnetlink_has_listeners(struct net *net, unsigned int group)
 118 {
 119         return netlink_has_listeners(net->nfnl, group);
 120 }
 121 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
 122
 123 struct sk_buff *nfnetlink_alloc_skb(struct net *net, unsigned int size,
 124                                     u32 dst_portid, gfp_t gfp_mask)
 125 {
 126         return netlink_alloc_skb(net->nfnl, size, dst_portid, gfp_mask);
 127 }
 128 EXPORT_SYMBOL_GPL(nfnetlink_alloc_skb);
 129
 130 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
 131                    unsigned int group, int echo, gfp_t flags)
 132 {
 133         return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags);
 134 }
 135 EXPORT_SYMBOL_GPL(nfnetlink_send);
 136
 137 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
 138 {
 139         return netlink_set_err(net->nfnl, portid, group, error);
 140 }
 141 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
 142
 143 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid,
 144                       int flags)
 145 {
 146         return netlink_unicast(net->nfnl, skb, portid, flags);
 147 }
 148 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
 149
 150 /* Process one complete nfnetlink message. */
 151 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
 152 {
 153         struct net *net = sock_net(skb->sk);
 154         const struct nfnl_callback *nc;
 155         const struct nfnetlink_subsystem *ss;
 156         int type, err;
 157
 158         /* All the messages must at least contain nfgenmsg */
 159         if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
 160                 return 0;
 161
 162         type = nlh->nlmsg_type;
 163 replay:
 164         rcu_read_lock();
 165         ss = nfnetlink_get_subsys(type);
 166         if (!ss) {
 167 #ifdef CONFIG_MODULES
 168                 rcu_read_unlock();
 169                 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
 170                 rcu_read_lock();
 171                 ss = nfnetlink_get_subsys(type);
 172                 if (!ss)
 173 #endif
 174                 {
 175                         rcu_read_unlock();
 176                         return -EINVAL;
 177                 }
 178         }
 179
 180         nc = nfnetlink_find_client(type, ss);
 181         if (!nc) {
 182                 rcu_read_unlock();
 183                 return -EINVAL;
 184         }
 185
 186         {
 187                 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
 188                 u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
 189                 struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
 190                 struct nlattr *attr = (void *)nlh + min_len;
 191                 int attrlen = nlh->nlmsg_len - min_len;
 192                 __u8 subsys_id = NFNL_SUBSYS_ID(type);
 193
 194                 err = nla_parse(cda, ss->cb[cb_id].attr_count,
 195                                 attr, attrlen, ss->cb[cb_id].policy);
 196                 if (err < 0) {
 197                         rcu_read_unlock();
 198                         return err;
 199                 }
 200
 201                 if (nc->call_rcu) {
 202                         err = nc->call_rcu(net->nfnl, skb, nlh,
 203                                            (const struct nlattr **)cda);
 204                         rcu_read_unlock();
 205                 } else {
 206                         rcu_read_unlock();
 207                         nfnl_lock(subsys_id);
 208                         if (rcu_dereference_protected(table[subsys_id].subsys,
 209                                 lockdep_is_held(&table[subsys_id].mutex)) != ss ||
 210                             nfnetlink_find_client(type, ss) != nc)
 211                                 err = -EAGAIN;
 212                         else if (nc->call)
 213                                 err = nc->call(net->nfnl, skb, nlh,
 214                                                    (const struct nlattr **)cda);
 215                         else
 216                                 err = -EINVAL;
 217                         nfnl_unlock(subsys_id);
 218                 }
 219                 if (err == -EAGAIN)
 220                         goto replay;
 221                 return err;
 222         }
 223 }
 224
 225 struct nfnl_err {
 226         struct list_head        head;
 227         struct nlmsghdr         *nlh;
 228         int                     err;
 229 };
 230
 231 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err)
 232 {
 233         struct nfnl_err *nfnl_err;
 234
 235         nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
 236         if (nfnl_err == NULL)
 237                 return -ENOMEM;
 238
 239         nfnl_err->nlh = nlh;
 240         nfnl_err->err = err;
 241         list_add_tail(&nfnl_err->head, list);
 242
 243         return 0;
 244 }
 245
 246 static void nfnl_err_del(struct nfnl_err *nfnl_err)
 247 {
 248         list_del(&nfnl_err->head);
 249         kfree(nfnl_err);
 250 }
 251
 252 static void nfnl_err_reset(struct list_head *err_list)
 253 {
 254         struct nfnl_err *nfnl_err, *next;
 255
 256         list_for_each_entry_safe(nfnl_err, next, err_list, head)
 257                 nfnl_err_del(nfnl_err);
 258 }
 259
 260 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
 261 {
 262         struct nfnl_err *nfnl_err, *next;
 263
 264         list_for_each_entry_safe(nfnl_err, next, err_list, head) {
 265                 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err);
 266                 nfnl_err_del(nfnl_err);
 267         }
 268 }
 269
 270 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
 271                                 u_int16_t subsys_id)
 272 {
 273         struct sk_buff *nskb, *oskb = skb;
 274         struct net *net = sock_net(skb->sk);
 275         const struct nfnetlink_subsystem *ss;
 276         const struct nfnl_callback *nc;
 277         bool success = true, done = false;
 278         static LIST_HEAD(err_list);
 279         int err;
 280
 281         if (subsys_id >= NFNL_SUBSYS_COUNT)
 282                 return netlink_ack(skb, nlh, -EINVAL);
 283 replay:
 284         nskb = netlink_skb_clone(oskb, GFP_KERNEL);
 285         if (!nskb)
 286                 return netlink_ack(oskb, nlh, -ENOMEM);
 287
 288         nskb->sk = oskb->sk;
 289         skb = nskb;
 290
 291         nfnl_lock(subsys_id);
 292         ss = rcu_dereference_protected(table[subsys_id].subsys,
 293                                        lockdep_is_held(&table[subsys_id].mutex));
 294         if (!ss) {
 295 #ifdef CONFIG_MODULES
 296                 nfnl_unlock(subsys_id);
 297                 request_module("nfnetlink-subsys-%d", subsys_id);
 298                 nfnl_lock(subsys_id);
 299                 ss = rcu_dereference_protected(table[subsys_id].subsys,
 300                                                lockdep_is_held(&table[subsys_id].mutex));
 301                 if (!ss)
 302 #endif
 303                 {
 304                         nfnl_unlock(subsys_id);
 305                         netlink_ack(skb, nlh, -EOPNOTSUPP);
 306                         return kfree_skb(nskb);
 307                 }
 308         }
 309
 310         if (!ss->commit || !ss->abort) {
 311                 nfnl_unlock(subsys_id);
 312                 netlink_ack(skb, nlh, -EOPNOTSUPP);
 313                 return kfree_skb(skb);
 314         }
 315
 316         while (skb->len >= nlmsg_total_size(0)) {
 317                 int msglen, type;
 318
 319                 nlh = nlmsg_hdr(skb);
 320                 err = 0;
 321
 322                 if (nlh->nlmsg_len < NLMSG_HDRLEN) {
 323                         err = -EINVAL;
 324                         goto ack;
 325                 }
 326
 327                 /* Only requests are handled by the kernel */
 328                 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
 329                         err = -EINVAL;
 330                         goto ack;
 331                 }
 332
 333                 type = nlh->nlmsg_type;
 334                 if (type == NFNL_MSG_BATCH_BEGIN) {
 335                         /* Malformed: Batch begin twice */
 336                         nfnl_err_reset(&err_list);
 337                         success = false;
 338                         goto done;
 339                 } else if (type == NFNL_MSG_BATCH_END) {
 340                         done = true;
 341                         goto done;
 342                 } else if (type < NLMSG_MIN_TYPE) {
 343                         err = -EINVAL;
 344                         goto ack;
 345                 }
 346
 347                 /* We only accept a batch with messages for the same
 348                  * subsystem.
 349                  */
 350                 if (NFNL_SUBSYS_ID(type) != subsys_id) {
 351                         err = -EINVAL;
 352                         goto ack;
 353                 }
 354
 355                 nc = nfnetlink_find_client(type, ss);
 356                 if (!nc) {
 357                         err = -EINVAL;
 358                         goto ack;
 359                 }
 360
 361                 {
 362                         int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
 363                         u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
 364                         struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
 365                         struct nlattr *attr = (void *)nlh + min_len;
 366                         int attrlen = nlh->nlmsg_len - min_len;
 367
 368                         err = nla_parse(cda, ss->cb[cb_id].attr_count,
 369                                         attr, attrlen, ss->cb[cb_id].policy);
 370                         if (err < 0)
 371                                 goto ack;
 372
 373                         if (nc->call_batch) {
 374                                 err = nc->call_batch(net->nfnl, skb, nlh,
 375                                                      (const struct nlattr **)cda);
 376                         }
 377
 378                         /* The lock was released to autoload some module, we
 379                          * have to abort and start from scratch using the
 380                          * original skb.
 381                          */
 382                         if (err == -EAGAIN) {
 383                                 nfnl_err_reset(&err_list);
 384                                 ss->abort(oskb);
 385                                 nfnl_unlock(subsys_id);
 386                                 kfree_skb(nskb);
 387                                 goto replay;
 388                         }
 389                 }
 390 ack:
 391                 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
 392                         /* Errors are delivered once the full batch has been
 393                          * processed, this avoids that the same error is
 394                          * reported several times when replaying the batch.
 395                          */
 396                         if (nfnl_err_add(&err_list, nlh, err) < 0) {
 397                                 /* We failed to enqueue an error, reset the
 398                                  * list of errors and send OOM to userspace
 399                                  * pointing to the batch header.
 400                                  */
 401                                 nfnl_err_reset(&err_list);
 402                                 netlink_ack(skb, nlmsg_hdr(oskb), -ENOMEM);
 403                                 success = false;
 404                                 goto done;
 405                         }
 406                         /* We don't stop processing the batch on errors, thus,
 407                          * userspace gets all the errors that the batch
 408                          * triggers.
 409                          */
 410                         if (err)
 411                                 success = false;
 412                 }
 413
 414                 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
 415                 if (msglen > skb->len)
 416                         msglen = skb->len;
 417                 skb_pull(skb, msglen);
 418         }
 419 done:
 420         if (success && done)
 421                 ss->commit(oskb);
 422         else
 423                 ss->abort(oskb);
 424
 425         nfnl_err_deliver(&err_list, oskb);
 426         nfnl_unlock(subsys_id);
 427         kfree_skb(nskb);
 428 }
 429
 430 static void nfnetlink_rcv(struct sk_buff *skb)
 431 {
 432         struct nlmsghdr *nlh = nlmsg_hdr(skb);
 433         int msglen;
 434
 435         if (nlh->nlmsg_len < NLMSG_HDRLEN ||
 436             skb->len < nlh->nlmsg_len)
 437                 return;
 438
 439         if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
 440                 netlink_ack(skb, nlh, -EPERM);
 441                 return;
 442         }
 443
 444         if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) {
 445                 struct nfgenmsg *nfgenmsg;
 446
 447                 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
 448                 if (msglen > skb->len)
 449                         msglen = skb->len;
 450
 451                 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
 452                     skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
 453                         return;
 454
 455                 nfgenmsg = nlmsg_data(nlh);
 456                 skb_pull(skb, msglen);
 457                 nfnetlink_rcv_batch(skb, nlh, nfgenmsg->res_id);
 458         } else {
 459                 netlink_rcv_skb(skb, &nfnetlink_rcv_msg);
 460         }
 461 }
 462
 463 #ifdef CONFIG_MODULES
 464 static int nfnetlink_bind(int group)
 465 {
 466         const struct nfnetlink_subsystem *ss;
 467         int type = nfnl_group2type[group];
 468
 469         rcu_read_lock();
 470         ss = nfnetlink_get_subsys(type);
 471         rcu_read_unlock();
 472         if (!ss)
 473                 request_module("nfnetlink-subsys-%d", type);
 474         return 0;
 475 }
 476 #endif
 477
 478 static int __net_init nfnetlink_net_init(struct net *net)
 479 {
 480         struct sock *nfnl;
 481         struct netlink_kernel_cfg cfg = {
 482                 .groups = NFNLGRP_MAX,
 483                 .input  = nfnetlink_rcv,
 484 #ifdef CONFIG_MODULES
 485                 .bind   = nfnetlink_bind,
 486 #endif
 487         };
 488
 489         nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
 490         if (!nfnl)
 491                 return -ENOMEM;
 492         net->nfnl_stash = nfnl;
 493         rcu_assign_pointer(net->nfnl, nfnl);
 494         return 0;
 495 }
 496
 497 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
 498 {
 499         struct net *net;
 500
 501         list_for_each_entry(net, net_exit_list, exit_list)
 502                 RCU_INIT_POINTER(net->nfnl, NULL);
 503         synchronize_net();
 504         list_for_each_entry(net, net_exit_list, exit_list)
 505                 netlink_kernel_release(net->nfnl_stash);
 506 }
 507
 508 static struct pernet_operations nfnetlink_net_ops = {
 509         .init           = nfnetlink_net_init,
 510         .exit_batch     = nfnetlink_net_exit_batch,
 511 };
 512
 513 static int __init nfnetlink_init(void)
 514 {
 515         int i;
 516
 517         for (i=0; i<NFNL_SUBSYS_COUNT; i++)
 518                 mutex_init(&table[i].mutex);
 519
 520         pr_info("Netfilter messages via NETLINK v%s.\n", nfversion);
 521         return register_pernet_subsys(&nfnetlink_net_ops);
 522 }
 523
 524 static void __exit nfnetlink_exit(void)
 525 {
 526         pr_info("Removing netfilter NETLINK layer.\n");
 527         unregister_pernet_subsys(&nfnetlink_net_ops);
 528 }
 529 module_init(nfnetlink_init);
 530 module_exit(nfnetlink_exit);